Building Hebrabic: one dataset, two languages, both directions
How a typed corpus of 1,617 words teaches Arabic to Hebrew speakers and Hebrew to Arabic speakers from the same data — with dialect layers, three interface languages, and an honesty policy for cognates.